Zhiyuan Tan1,2, Aymara Sancho-Araiz1, Swantje Völler1, SofÃa Peeters2,3, Maarten J. Deenen2,3, Jan Gerard Maring4, Pierre M. Bet5,6, Ron A. A. Mathôt5, Catherijne A. J. Knibbe1,7, Dirk Jan A. R. Moes2
1Division of Systems Pharmacology and Pharmacy, Leiden Academic Centre for Drug Research, Leiden University, 2Department of Clinical Pharmacy & Toxicology, Leiden University Medical Center, 3Department of Clinical Pharmacy, Catharina Hospital Eindhoven, 4Department of Clinical Pharmacy, Isala Hospital Zwolle, 5Department of Hospital Pharmacy and Clinical Pharmacology, Amsterdam University Medical Centre, 6Cancer Center Amsterdam, 7Department of Clinical Pharmacy, St Antonius Hospital
Objectives: 5-Fluorouracil (5-FU) is the cornerstone of first-line chemotherapy for various gastrointestinal cancers[1]. However, up to 76% of the patients treated with 5-FU based chemotherapy experiences severe toxicity[2] depending on the type of treatment regimen. Recently, a narrow therapeutic index for the infinitive area under the curve (AUC0-inf) target of 20–30 mg·h/L was proposed, that is reached by only 20.3% of the patients[3]. Therefore, dose individualization through model-informed precision dosing (MIPD) is warranted thereby optimizing therapeutic outcomes and improving patient safety. Here, we collected various prospective individual pharmacokinetic (PK) datasets to externally evaluate existing population pharmacokinetics (POPPK) models and further establish a more generic, adaptable model to accurately predict individual 5-FU exposure across diverse regimens. Additionally, optimal limited sampling strategies (LSS) were identified to estimate individual AUC0-inf and a MIPD shiny application was developed to facilitate dose adjustment algorithms. Methods: Patients from four prospective clinical studies were included with information on demographic data (age, sex, BSA, BMI, diagnoses, etc.), dosing schedule, PK data, and DPYD activity score. A systematic evaluation of existing 5-FU POPPK models[4-9] was conducted to inform PK modelling. The published models were externally evaluated with maximum a posterior Bayesian estimation, and subsequent PK modeling was performed with covariate analysis where required. Both NONMEM $DESIGN and manual screening of possible LSS were evaluated. Simulation of the AUC0-inf was performed to generate initial dosing strategies for patients. A user-friendly shiny application was developed to enable subsequent dose adjustments. Results: From 140 individuals (42.1% colorectal cancer and 34.3% pancreatic cancer), 435 PK samples were included. The median age was 63 [IQR 56-70] years, median body weight was 79 [IQR 65-89] kg, and median BSA was 1.94 [IQR 1.75-2.12] m². While 14 patients received a 24-hour infusion, the other patients received a 46-hour infusion with or without a prior 5–15 minute bolus. Using the published models, population predictions resulted in underestimation of observed concentrations and misspecification of 24-hour infusion regimen was observed, with two-compartment Michaelis-Menten model outperforming the other model structures. In the developed POPPK model, concentrations were best described using a two-compartment model with Michaelis-Menten elimination. Inter-individual variability (IIV) on the maximum rate of reaction (Vmax) was identified, and BSA partly explained the IIV, with an estimated allometric exponent of 1.14 (IIV reduction of 12%). The optimal LSS for bolus and continuous infusion of 46 hours was T= 0.1 (end of bolus), 1, and 3 hours post-administration, with a mean absolute percentage error (MAPE) of 5%. For the 46-hour continuous infusion without bolus, the optimal sampling times were 2, 3, and 5 hours, with a MAPE of 8.5%. The standard dose of 400 mg/m² bolus + 2400 mg/m² continuous infusion achieved a median AUC0-inf of 29 mg·h/L (90% Prediction Interval, PI: 17–50 mg·h/L). To improve target attainment, the bolus dose could be reduced to 200 mg/m² to achieve a median AUC0-inf of 24 mg·h/L (90% PI: 14–41 mg·h/L). For continuous infusion regimen, the standard dose 2400 mg/m² resulting in a median AUC0-inf of 21 mg·h/L (90%PI: 12–36 mg·h/L) could be increased to 2800 mg/m² resulting in a median AUC0-inf of 25 mg·h/L (90% PI: 14–43 mg·h/L). A shiny application was developed to provide dose recommendations for subsequent cycles based on measured concentrations from the previous cycle (https://tanzy1995.shinyapps.io/5FU-shiny-app/). Conclusions: The external evaluation of 5-FU POPPK models against a pooled prospective population with diverse cancer types and dosing regimens resulted in a generic two-compartment Michaelis-Menten PK model with BSA as covariate on Vmax. Based on these results, it is proposed to decrease the initial dose to 200+2400 mg/m2 for bolus and continuous infusion, while for continuous infusion only, the initial dose could be increased to 2800 mg/m2. The utilization of LSS together with MIPD shiny application allows for precise estimation of 5-FU AUC0-inf in routine clinical practice, and may provide dose individualization recommendations that reduce severe toxicities without compromising efficacy.
[1] Lee J et al. Cancer Chemother Pharmacol (2016) 3, 447-64. [2] Conroy T et al. N Engl J Med (2018) 25, 2395-2406. [3] Beumer J et al. Clin Pharmacol Ther (2019) 3, 598-613. [4] Bressolle F et al. Cancer Chemother Pharmacol (1999) 4, 295-302. [5] Muller F et al. Cancer Chemother Pharmacol (2013) 2, 361-370. [6] Terret C et al. Clin Pharmacol Ther (2000) 3, 270-279. [7] van Kuilenburg, A. B. P et al. Clinical Pharmacokinetics (2012) 3, 163-174. [8] Woloch C et al. Current topics in medicinal chemistry (2012) 15, 1713-1719. [9] PAGE 32 (2024) Abstr 10998.
Reference: PAGE 33 (2025) Abstr 11695 [www.page-meeting.org/?abstract=11695]
Poster: Clinical Applications